[GE users] pe_slots issues

Reuti reuti at staff.uni-marburg.de
Fri Nov 2 16:19:35 GMT 2007


Am 02.11.2007 um 16:13 schrieb John Coldrick:

> On Friday 02 November 2007 10:52, Reuti wrote:
>> Am 02.11.2007 um 14:32 schrieb John Coldrick:
>>> On Thursday 01 November 2007 17:43, Reuti wrote:
>>>> Am 01.11.2007 um 21:16 schrieb John Coldrick:
>>>>> 	I've got a PE set up:
>>>>> ***
>>>>> 	pe_name           m9_1
>>>>> 	slots             999
>>>>> 	user_lists        NONE
>>>>> 	xuser_lists       NONE
>>>>> 	start_proc_args   /bin/true
>>>>> 	stop_proc_args    /bin/true
>>>>> 	allocation_rule   $pe_slots
>>>>
> ...
>> So only one queue on these 8-core machines?
>
> 	I have only one queue defined, the default all.q, and every system  
> on the
> grid has an instance on it, 8 core instances included.
>

Good.

>
>>>> Often advisable with parallel jobs is to request reservation  
>>>> with "-R
>>>> y" in qsub and set a sensible value for "max_reservation" in the
>>>> scheduler configuration.
> ...
>> It's the number of jobs. 20 maybe good.
>
> 	Ah, OK.  Doesn't seem to affect this issue.
>
>> In the queue definition you defined more slots for these three  
>> machines?
>
> 	Yes, each system has slots equal to the number of cpus.  If I  
> submit the same
> jobs single threaded(i.e. don't use the PE m9_1 at all), they all  
> run on
> every slot(e.g. those 8 core machines have 8 jobs running on  
> them).  This
> seems to be specifically about the PE.
>
> 	The only thing I've configured in this fresh install of 6.1u2  
> apart from a
> few complexes to manage licenses and adding the PE  is this:
>
> 	I've changed the existing complex mem_free to be consumable.  I  
> define each
> system to have whatever specific memory they actually have and use  
> this
> system to help manage memory e.g. if a job needs 4G to run, it  
> requests that
> and SGE manages it nicely.  To ensure this isn't at the root of the  
> problem,
> I've just tested now with mem_free being non-consumable(the  
> default), and it
> still doesn't work.

Resource requests of consumables will be multiplied by the number of  
requested slots for parallel jobs.

-- Reuti


> 	Thanks
>
> 	J.C.
>
> -- 
> John Coldrick                  www.axyzfx.com        Axyz Animation
> 416-504-0425                                         477 Richmond St W
>                                                      Toronto, ON  
> Canada
> jc at axyzfx.com                                        M5V 3E7
> ---------------------------------------------------------------------- 
> -
> "Life is too important to take seriously."
> 		-- Corky Siegel
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list